This study empirically evaluates four infrastructure characteristics of container image size, startup time, idle resource consumption, and horizontal scaling latency to yield context-dependent selection guidance and a fully reproducible benchmark suite.
Abstract
Framework and runtime selection for containerized microservices are usually guided by request-level benchmarks, yet deployment-facing costs often dominate operational expenditure in Kubernetes environments. This study empirically evaluates four such infrastructure characteristics: container image size, startup time, idle resource consumption, and horizontal scaling latency. Eight microservice framework configurations spanning three execution models are evaluated: JVM (Spring Boot, Spring WebFlux, Quarkus, and Ktor), GraalVM Native image (Quarkus variants, including distroless and UPX-compressed images), and Rust (Actix Web). All metrics are collected technology-agnostically at the container level via cAdvisor and Kubernetes lifecycle events. Three trade-off profiles emerged during the research: Rust achieves a 2.95 MiB idle memory footprint (a 69:1 ratio versus Spring Boot on a working-set basis, or 21:1 on the more conservative proportional-set-size basis) through garbage-collector-free memory management. GraalVM Native image variants start 1.2–1.7× faster and consume up to 1.9× less memory than their JVM equivalents, although this memory advantage is not uniform: the standard reactive Native image consumes more idle memory (115.0 MiB) than the corresponding JVM variant (98.0 MiB). A UPX compression paradox is identified and explained at the kernel level: compression shrinks images by approximately 2.5:1 yet inflates idle memory to 215–229 MiB, above JVM baselines, because decompression into private anonymous memory defeats shared page mapping. Scale-up latency (1.8–3.9 s) is governed by per-instance startup rather than framework-exclusive lifecycle optimizations, partially refuting one of four research hypotheses. The findings yield context-dependent selection guidance and a fully reproducible benchmark suite. All measurements were obtained on a single-node bare-metal K3s cluster, the primary metrics characterize the idle state of a minimal no-operation service, and the only load applied is a single fixed-rate validity check at 100 requests per second. The reported values therefore constitute lower-bound, deployment-facing infrastructure costs rather than predictions of behavior under production business workloads, multi-node topologies, or managed cloud substrates.
AddressSanitizer (ASan) is widely deployed in industrial C/C++ development, but its runtime and memory costs can limit routine use in continuous integration and production-like testing. Although many optimized ASan-like sanitizers have been proposed, heterogeneous workloads, configurations, and baselines make their pra...
Jia-Xin Xi, Cheng Wen, Zhi-Wei Lin et al.· Proceedings of the 41st IEEE...· 0 citations
WebAssembly (Wasm) has emerged as a lightweight, portable, and secure execution environment beyond the browser, with growing adoption in data-intensive, machine learning, and performance-sensitive workloads. However, the runtime ecosystem remains fragmented, and a systems-level understanding of how runtime design affec...
Sallar Khan, Tania Malik, K. Hasanov· IEEE Access· 0 citations
Kubernetes 1.35 stabilized in-place Pod resource resizing, but a deployable controller must avoid duplicate actuation, decline infeasible changes safely, and make blocked actions distinguishable from controller failure. This paper presents Elastic Pod CPU Auto-Scaler (EPCAS), an open-source operator for restart-sensiti...
H. Rajagopal, Kavya Vishwanath, Daniel Shafer et al.· IEEE Access· 0 citations
Empirical evaluation on NVIDIA V100 and AMD MI210 hardware shows Baseliner reaches measurement parity with vendor libraries within the precision floor of GPU event timing while remaining competitive in wall-clock time.
C. Eyraud, Maxime Lamothe· Proceedings of the 41st IEEE...· 0 citations
This paper investigates strategies for optimizing memory management and minimizing GC overhead in cloud-native Java applications, and analyzes modern garbage collection algorithms, evaluate native compilation via GraalVM, explore the memory impact of Virtual Threads, and formulate container-aware tuning strategies for...
Tulsiram Sharma, Shubham Agrawal· International Journal of Res...· 0 citations
AdapBench is the first approach which dynamically makes at-runtime decisions for prioritizing the most informative benchmarks and early-stopping each one once enough evidence has accumulated to accept or reject a performance change hypothesis.
Sebastian Koch, Jonas Heisterberg, Nils Japke et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.